Action Level Prediction on Abg-CoQA (OOD)
93.44Ambiguous AccProMed(S#1)+DPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ProMed(S#1)+DPOTraining=CMB-trained, Backbone=LLaMA-8B, Method Variant=S#1 + DPO2025.08 | 93.44 | 9.11 | 51.28 | |
| ProMed(S#1+2)Training=CMB-trained, Backbone=LLaMA-8B, Method Variant=S#1+22025.08 | 79.51 | 34.3 | 56.91 | |
| ProMed(S#1)+GRPOTraining=CMB-trained, Backbone=LLaMA-8B, Method Variant=S#1 + GRPO2025.08 | 77.24 | 30.93 | 54.08 | |
| VanillaTraining=CMB-trained, Backbone=LLaMA-8B2025.08 | 60.16 | 48.11 | 54.15 |